PUE: Making Friends with the Monster in the Cupboard
There are many discussions regarding PUE (power usage effectiveness) nowadays. While more and more people want to at least have an idea of what PUE they are running at, there are quite a few who think they already know. There are also companies offering to improve your PUE and ask you to invest money over a certain amount of years. All of those efforts are fantastic. However, there is a monster in the cupboard that some will not address, or at least, will try not to address. The name of that monster is “Power Factor”. Our Head of Infrastructure, Pavelas Sokolovas, discusses.
So, what is Power Factor?
At any given moment, it is the amount of power being used in Watts (or Kilowatts, Megawatts etc.) divided by Volt-Amps (or Kilo-VAs, Mega-VAs etc.). As we remember from school, power is equal to voltage multiplied by current. So, P=VA, in other words, 1W=1VA.
Where is this monster?
The monster is when we have an alternating current (AC) and not a direct current (DC) system, and the load is not resistive. If the load is 100% resistive, then the above equation is correct where 1W=1VA despite the load being fed from AC supply. We are using so-called RMS (root mean square) values when calculating the current and voltage in an AC environment. RMS – in short – is a value of equivalent DC power supply that would provide exactly the same amount of energy over one single period of an alternating current on the same resistive load. Therefore, it means when we are talking about 230V AC supply, that the peak voltage is not 230V, but as high as 325V. This makes RMS value of 230V, the same principle that applies to the current.
The monster gets out of its cupboard and starts to roar when the load is not 100% resistive. The main load of a data centre is switching power supplies. Most of them have an AC/DC bridge and a large capacitor facing the power feed. Therefore, it is a capacitive load. Whereas most air conditioners, chillers and other infrastructure elements are all inductive loads. This is because they have big coils facing the power feed.
Capacitive loads create a negative power factor and inductive loads create a positive power factor. It has nothing to do with being good or bad. In the first scenario, the current is leading voltage to charge the capacitor. In the second scenario, the current is always lagging because of autoinduction phenomena of the coil.
Therefore, if we connect a massive capacitor to AC power supply, there will be current while it is charging and discharging. Consequently, voltage multiplied by current will be above zero. Interestingly, the power usage in Watts will be zero on average; simply because the capacitor is giving its stored energy back to the grid just like a coil (the only active energy used in this process will be a power loss on the resistive component of the coil).
How PUE is measured
There are several ways to measure PUE and they will give you the same result, providing you do not make any mistakes. You can:
- Measure the IT load in Watts and divide the total facility power by it
- Measure non-IT load in Watts; subtract it from total facility power in Watts (which will give us an IT load) and divide the total facility load by IT load calculated.
There are more ways than this; the one you choose depends on what is more convenient to measure. For example, you can use (1) if you have a reliable way of measuring the IT load in Watts. In case you do not have a convenient way to measure the IT load in Watts but you do have it to measure the non-IT load in Watts, then you can use (2). The key is to use the correct units (Watts, not Amps or VAs) and to avoid any mistakes in the formula (e.g. forget to add a power measurement of one or multiple racks or infrastructure components or to add them multiple times).
My brother is a mathematician and I remembered he could “prove” the following on paper. He “proved” that two is equal to three. Despite my quite strong math knowledge, it was not easy to spot the trick. It was one of the things where you should change a sign when going square, square root or/and modulus. My point is – if the result would not appear so obviously wrong – then I would not spot the mistake in thinking that two is equal to three.
Interestingly, when calculating PUE, almost everyone performs it using the first method. Most of them do it because there are lots of measuring tools already in place. For example, each rack of IT load has a metered PDU (power distribution unit) for the billing and/or monitoring purposes as opposed to fans, air conditioners and the rest of the infrastructure equipment. What they do is add all of them up to get the entire IT load. Then they take the entire facility load and divide it by the total IT load calculated.
However, this is where the monster begins to laugh. The reason being: they have used the wrong units. Most of rack PDU’s measure Amps, but as we know, Amps multiplied by Volts does not give us power in Watts, but VAs, which is not equal to Watts (and it can only be higher). In addition, most rack PDU’s have an accuracy of 1A or 10% (whichever is greater) so you can imagine the potential inaccuracy of using this method!
I can forgive the majority for doing that because they are unaware of this “monster”, but I cannot forgive anyone who does it on purpose to hide the real figure. This is simply not fair play. As modern power supplies have a power factor of 0.85-0.95, if we were to measure only current (Amps) and multiply the total by voltage, we would get up to 1.18 times higher values than they actually are.
A simpler example is a 230V data centre, which uses 1MW of power, while IT load is 700kW and 300kW of this being used by air conditioners, fans and other infrastructure equipment. In this case, we would have a PUE of 1.43. Now imagine you want to use the first method to calculate PUE. If you total all Amps from all racks, and your load power factor is 0.85, you would get 3580 Amps.
Remember, at this moment you do not know (or are yet to be made aware of) the power factor. You assume that if you multiply your Amps by 230V that you will get IT load (what you are actually believing here is that two equals three). So the calculation 3580 x 230 = 823.4kW, if we take this to calculate the PUE we will get 1MW/823.4kW = 1.22.
The “monster” is the Power Factor. If you consider this, your “PUE” will be 1.22 instead of the real 1.43. Quite a difference! Also, the more readings you total, the bigger the combined error. If each PDU has a 0.1A error, and your average rack power usage is 10A, you can get 1% error in calculating total Amps. You can also forget to add some of them or add some of them multiple times. It is very easy to make this mistake because there are so many of them! In addition, what you may not know is PDUs with 1% accuracy are rare and expensive. You will most probably have 1A or 10% (whichever is higher) accurate PDUs installed. Be sure not to think that just because they are from reputable brands, that they will have this accuracy.
At Custodian Data Centres, we decided to use method two when measuring PUE. We installed accurate and powerful meters to measure Watts everywhere except for IT load. This included the site 11kV supplies, after every UPS and at each distribution board. We separated IT load distribution boards from all other infrastructure distribution boards, so we can deduct the correct amount of Watts used by infrastructure, and subtract it from the incoming power so that we get the exact figure for IT load. We then divided the total power used, by total power used minus power used by infrastructure.
It is as easy as that. What’s more, it is a lot more precise. Our PUE is still at an impressive 1.2 on average and we are (to my best knowledge) the only colocation facility to display our PUE live on our website. Having to add and subtract only several readings, we reduce the error to a minimum. We can proudly claim that our PUE really is 1.2. For example, if we were to use the first method of calculation mentioned (and used Amps) our calculated “PUE” would be 1.06 or even less than one (which is impossible). That would be as obviously wrong as two equal three, would it not?
A bit of advice for data centre owners: although many PUE “improvers” really can improve your PUE, it does not mean they will improve it to the level they say. When signing a contract and performing an audit, please make sure that you always measure Watts (not VAs) for the purpose of the PUE calculation.
Then the monster in the cupboard will be your smiling friend forever!